ContentFlow AI

Home Screen
Input Processing
Output

Inspiration

Content creation today is either time-consuming or generic. While large language models can generate text quickly, they often lack structure, research depth, and coherence across long-form content.

The idea behind ContentFlow AI was to go beyond simple prompting and build a system that mimics how humans actually write: → plan → research → organize → write → refine

This led to the concept of using agentic workflows, where multiple AI agents collaborate to produce high-quality, structured blog content automatically.

What it does

ContentFlow AI is an AI-powered multi-agent content engine that transforms a simple topic into a fully structured, research-backed blog.

It:

Generates a content plan with sections and structure
Performs research (optional) to gather relevant information
Uses multiple agents to write different sections in parallel
Merges outputs into a coherent final blog
Provides a clean Markdown output ready for publishing

The system ensures that content is not just generated-but thought through and assembled intelligently.

How we built it

The project is built using a modular agent-based architecture:

LangGraph → orchestrates the workflow as a stateful graph LangChain → handles LLM interactions and tool integration Groq API → powers fast LLM inference Streamlit → provides an interactive frontend UI

Workflow Design The system follows a structured pipeline:

Router → Research → Orchestrator → Workers → Reducer

Router Node: Decides whether research is required Research Node: Fetches external knowledge (if enabled) Orchestrator Node: Creates a detailed content plan Worker Nodes: Generate sections in parallel Reducer Node: Combines everything into final output

This architecture allows parallel execution and scalability, making it efficient and modular.

Challenges we ran into

Dependency conflicts during deployment Managing compatibility between libraries (like numpy, streamlit, langchain) required careful version handling.
Designing agent coordination Ensuring that multiple agents produce consistent and non-redundant content was tricky.
Handling optional research flows Dynamically switching between closed-book and research-based generation added complexity.
UI/UX limitations in Streamlit Making the interface intuitive across devices (especially mobile) required thoughtful design decisions.
Environment variables & API keys Managing API keys securely during deployment (e.g., Groq, Tavily) was a key learning point.